Background. Emergency settings demand rapid, accurate decisions under crowding and resource strain. Artificial intelligence (AI) has been proposed to enhance triage, diagnosis, and early deterioration prediction across prehospital and emergency department (ED) workflows. Objective. To synthesize contemporary evidence on AI applications in medical emergency situations, integrating original studies and recent systematic reviews. Methods. We reviewed 19 recently uploaded open-access studies: 10 original investigations spanning triage optimization, diagnostic support, and arrest prediction, and 9 systematic/scoping reviews summarizing ED and prehospital AI. We extracted design, inputs, models, validation, and clinical performance (AUROC/AUPRC, sensitivity/specificity, mis-triage). Results. AI consistently matched or outperformed conventional tools across tasks. Machine-learning triage reduced mis-triage (prospective 0.9% vs 1.2%) and achieved AUROC 0.875, while level-3 streaming models reached AUROC 0.755–0.761. Deep models predicted in-hospital cardiac arrest from ECG (AUROC 0.913–0.948) and multimodal ED data (AUROC 0.904–0.939). Emergency radiology AI detected intracranial hemorrhage with sensitivity 88.8% and boosted combined reader+AI sensitivity to 95.2%. NLP on EMS notes improved prehospital stroke identification (c-statistic 0.73 vs 0.53–0.67 for rules). Conclusions. Evidence supports AI as decision support for triage, arrest prediction, and imaging in emergencies, with strongest gains from gradient-boosting and deep learning using structured vitals, ECG, text, and imaging. Key gaps include external validation, workflow integration, fairness, and patient-centered outcomes.